Documentation

Pellucid documentation

Everything you need to detect ambiguity in your documents, plug Pellucid into your stack, and run it on your own infrastructure.

Pellucid is the clarity layer for high-stakes writing — software requirements, contracts, compliance policies. Two layers do the work: a deterministic rule layer that fires in about 200 ms, and a debating multi-agent layer that adds context-sensitive findings over the next 5–15 seconds. Both produce the same Finding shape, so any consumer (web app, browser extension, VS Code, your own pipeline) can render the result identically.

Start here

Pick the entry point that matches what you’re trying to do. Every guide below stands alone — you don’t need to read them in order.

Guides

Quickstart

Paste a doc, watch the rule layer fire, then watch four agents debate. Five minutes from cold to first rewrite.

Guides

Bring your own key

Pellucid never bakes in a provider. Plug in xAI, Anthropic, OpenAI, or OpenRouter — env var or per-install setting.

Reference

API reference

Every shipped endpoint with cURL, Python, and JS examples. Documents, findings, debate, share, ontology, glossary, templates, rules, audit, analytics.

Reference

Webhooks

analysis.completed payload, HMAC-SHA256 signing, and copy-paste verification snippets in Python and Node.

Extensions

Browser & VS Code

Install the Chrome extension and the VS Code extension. Right-click any selection or annotate a markdown file inline.

Operate

Self-hosting

Local dev, Docker Compose for Postgres + Neo4j, and notes for shipping Pellucid behind your own domain.

What’s shipped today

The Session 1 build covers the full clarity loop. Anything not listed here is on the roadmap — the docs only describe what actually runs.

Rule layer. spaCy-backed detectors for vague terms, passive voice, ambiguous quantifiers, pronoun antecedents, and a handful of other INCOSE-derived patterns. User-authored regex rules layer on top.
Multi-agent layer. Four specialists (Lexical, Syntactic, Domain, Risk) run in parallel under LangGraph, then critique each other in a single round. The aggregator merges spans and weights confidence by agent agreement.
Forge Clarity rewrites. Pick any finding and get 2–3 candidate rewrites with rationales — the same model that produced the finding produces the rewrite.
BYOK. Default model is xAI Grok 4. Anthropic, OpenAI, and OpenRouter work with no code changes — set an env var or paste a key in/settings.
Integrations. Outbound webhooks (generic JSON, Slack, Discord) fire on analysis completion, signed with HMAC-SHA256. Mintable API keys for future inbound auth.
Audit + share. Download a JSON or Markdown audit trail per document. Share a document publicly with a tokenised read-only URL. Embed a clarity badge in your README.
Extensions. A Chrome/Edge browser extension that scans highlighted text from any page; a VS Code extension that surfaces findings as native diagnostics with quick-fix rewrites.

Conventions used in these docs

Code samples. Every API endpoint includes cURL, Python (httpx), and JavaScript (fetch) tabs. Copy with the button in the bottom-right of any block.
Base URL. Examples assume http://localhost:8000. Replace with your deployment’s origin.
Auth. The Session 1 API runs unauthenticated by default. The API reference notes which endpoints will accept the bearer keys minted via /api/v1/integrations/api-keys when the auth middleware lands.
Versioning. All endpoints sit under /api/v1. Webhook payloads carry a schema_version field; bumping it is a breaking change.

Something missing?

If a behaviour you depend on isn’t documented, that’s a bug — open an issue and we’ll either write it up or remove the surface. These docs only describe the parts of Pellucid that are actually shipped and load-bearing.